Question 1 [13 marks]

Select the correct answer and provide a brief explanation:

1. [2 marks] The figure below shows a training set of 8 examples described with one numerical feature x and belonging to two classes: circles and squares. A new example is shown with a blue triangle.

What will be the prediction of 3-Nearest Neighbor for the class of the new example? If

there are ties, settle them by choosing the example on the left.

Circle                  Square

Explanation:

2. [2 marks] Does this diagram correspond to Bagging or Boosting?

Bagging Boosting

Explanation:

3. [3 marks] The kernel trick in support vector machines ensures that the data will be linearly separable in the new space.

True                    False

Explanation:

4. [3 marks] Given is the following training data, where occupation, age and loan-salary- ratio are the features and outcome is the class. Two prediction models are built, Model 1 and Model 2, both consistent with the training data.

Question 2 [10 marks]

Given is the following training data, where city and season are the features and price is the class:

Use Naïve Bayes to predict the value ofprice for the following new example: city=Sevilla, season=summer. Show your calculations.

Question 3 [10 marks]

Given is the following training data, where restaurant and time are the features and price is the class:

You may use this table:

a)What is the entropy of this data set with respect to the class?

b) What is the information gain of restaurant? Show your calculations.

Question 4 [13 marks]

1. [4 marks] There are 100 students in a computer science course. Isabella consistently outperforms the other students on the assessments during the semester and on the final exam he gets a mark of 99 while the next highest mark is 75. The range of exam marks is between

5 and 99. We would like to fit a linear regression model to the exam marks. Would Isabella’s mark cause problems? Briefly explain your answer.

2. [2 marks] List one advantage of Lasso regression compared to the standard linear regression and briefly explain your answer.

3. [2 marks] In random forest, how is the correlation among the combined decision trees reduced?

4. [5 marks] Consider the task of predicting credit card fraud in real time using a machine learning classifier. This task requires that the classifier performs thousands of predictions per second. Which algorithm is more suitable: k-nearest neighbor or logistic regression? Explain your answer.

Question 5 [14 marks]

1.  [8 marks] A company is building a classifier to predict if customers will like new products. The classifier takes as an input a vector with a very high dimensionality, has to be trained on a very large dataset and also has to